Room #308
Hybrid format - See details here.
ICML Website for registrants: https://icml.cc/virtual/2022/workshop/13460
Distribution-free methods enable rigorous uncertainty quantification with any (misspecified) model and (unknown) data distribution.
Accuracy alone does not suffice for reliable, consequential decision-making; we also need uncertainty.
Distribution-free UQ gives finite-sample statistical guarantees for any predictive model, no matter how bad/misspecified, and any data distribution, even if unknown.
DF techniques such as conformal prediction represent a new, principled approach to UQ for complex prediction systems, such as deep learning.
This workshop will bridge applied machine learning and distribution-free uncertainty quantification, catalyzing work at this interface.
Distribution-free methods make minimal assumptions about the data distribution or model, yet still provide uncertainty quantification. Examples of DF methods include conformal prediction, tolerance regions, risk-controlling prediction sets, calibration by binning, and more. We take a broad outlook on DF methods, so any assumption-light uncertainty quantification approaches are welcome.